Basic classification examples

Prerrequisites

library(tidyverse)
library(dslabs)
library(reshape2)
library(patchwork)
library(tidyquant)

Digits classification

Load the data.

mnist <- read_mnist()
glimpse(mnist)

List of 2
 $ train:List of 2
  ..$ images: int [1:60000, 1:784] 0 0 0 0 0 0 0 0 0 0 ...
  ..$ labels: int [1:60000] 5 0 4 1 9 2 1 3 1 4 ...
 $ test :List of 2
  ..$ images: int [1:10000, 1:784] 0 0 0 0 0 0 0 0 0 0 ...
  ..$ labels: int [1:10000] 7 2 1 0 4 1 4 9 5 9 ...

Save the features into a different variable.

features <- mnist$train$images
glimpse(features)

 int [1:60000, 1:784] 0 0 0 0 0 0 0 0 0 0 ...

Converting our data to matrix.

d <- matrix(features[1,], nrow = 28, byrow = TRUE)
digito <- melt(d)
digito

We plot the resulting image.

digito %>% 
  ggplot(aes(x = Var2, y = Var1)) +
  geom_raster(aes(fill = factor(value))) +
  scale_y_reverse() + scale_fill_grey() +
  theme(legend.position = "none")

p <- function(i){
    d <- matrix(features[i,], nrow = 28, byrow = TRUE)
    digito<-melt(d)
    ggplot(digito, aes(x = Var2, y = Var1)) +
    geom_raster(aes(fill=value)) +  coord_fixed() +
      theme_void() + theme(legend.position = "none") +
      scale_y_reverse()
  }

p(1) + p(2) + p(3) + p(4) + p(5) + p(6)

We get the total occurences for each digit.

etiquetas <- mnist$train$labels 
etiquetas %>%
  as_tibble() %>% 
  mutate(cant = 1) %>%
  group_by(value) %>%
    summarise(total = sum(cant))

`summarise()` ungrouping output (override with `.groups` argument)

d %>% 
  as_tibble()

features2 <- features %>% 
  as_tibble() %>% 
  mutate(digito = etiquetas,
         n = 1:60000) %>% 
  select(n, digito, everything())
features2

features_long <- features2 %>% 
  pivot_longer(cols = -c(n, digito)) %>% 
  mutate(Var1 = rep(1:28, times = 28 * 60000),
         Var2 = rep(1:28, each = 28, times = 60000)) %>%
  select(-name)
features_long

We create a function that selects a subset of the data and plots the result.

For example, taking the first 12 rows.

p <- function(i){
  features_long %>% 
    filter(n %in% i) %>%
    ggplot(aes(x = Var1, y = Var2)) +
    geom_raster(aes(fill = factor(value))) +
    scale_y_reverse() + scale_fill_grey() +
    theme(legend.position = "none") + 
    facet_wrap(~ n)
}

p(1:12)

tibble(Var1 = 3,
                  Var2 = 6,
                  lab = features_long %>% group_by(n) %>% distinct(digito) %>% pull(digito)) %>% 
  slice(2:4)

Getting now the second 12 rows:

mnist_clustering <- kmeans(features, centers = 10, nstart = 10)

did not converge in 10 iterationsQuick-TRANSfer stage steps exceeded maximum (= 3000000)Quick-TRANSfer stage steps exceeded maximum (= 3000000)Quick-TRANSfer stage steps exceeded maximum (= 3000000)

glimpse(mnist_centers)

 num [1:10, 1:784] 0 0 0 0 0 0 0 0 0 0 ...
 - attr(*, "dimnames")=List of 2
  ..$ : chr [1:10] "1" "2" "3" "4" ...
  ..$ : NULL